Cocktail algorithm studies MB Workshop Cocktail algorithm studies Carmen Diez Pardos Silvia Goy López CIEMAT Madrid Muon POG 07/04/2011 C. Diez Pardos 1 C. Diez Pardos (CIEMAT) 1
Introduction: cocktail algorithm Study in detail chi2, resolution and pulls for the data and MC used in the W' analysis Not many high pt muons Also: compare with cosmics (next step) There are several tunes, the used one is “TuneP” Algorithm logic: selection based on -lnProbchi2 Picky is taken as default, if it doesnt exit -> Take TPFMS -> tracker - >global If -lnProbchi2 (Picky-Tracker) >30, chosen tracker If -lnProbchi2 (TPFMS-chosen) >0 : takes TPFMS C. Diez Pardos 2 2
Samples and Muon Selection Data: 2010RunB Nov4ReReco MC: Fall10 W samples, W’ (1500GeV) Selection on the criteria for the analysis Must be Global and Tracker Muons Combined isolation: < 0.15 in a cone R < 0.3 Quality cuts related to the track: d0<0.2 cm, number of valid muon and pixel hits >0, number of valid tracker hits >10, number of matching segments >1, valid muon hits >0 Muon matched to a L3 muon: HLT_Mu9, _Mu11, _Mu15 Note: For this selection W MC is not enough to describe data (should include other BG), but it should be fine for pT > 100 GeV. C. Diez Pardos 3 3
Cocktail choice for data and MC At low pT data and MC differ in Barrel and EC: Low pt W: more TPFMS Data: more Picky (changes tendency at high pT, see next slide) TkOnly contibrutes ~1% C. Diez Pardos 4 In general, TPFMS preferred, for data Picky in the Barrel until pt 200 4
Cocktail choice for data MORE TPFMS ACORDING TO joRDAN
-LnProb(chi2) difference TPFMS- PICKY Negative mean! BARREL W W' W ENDCAP W' Mass 1500 GeV C. Diez Pardos 6 6
Ratio chi2 MC W over Data pt>100 GeV In this region most of the BG is W, distributions are normalised to area C. Diez Pardos 7 7
Ratio ln chi2 tail prob MC W over Data pt>100 PICKY TPFMS In this region most of the BG is W, distributions are normalised to area C. Diez Pardos 8 8
Pt distributions ANIADIR RATIO, PT for pt<100 W C. Diez Pardos 9 9
Resolution residuals for the chosen algo Fit for each algorithm the resolution, for the whole region and the BARREL ONLY?? PUT a plot C. Diez Pardos 10 10 Left tail: reconstructed momentum higher than generated Interested: See tails in low pt distributions and resolution core in high pt MEAN: Similar? SIGMA: Quite similar within algorithm, better for the chosen one
Resolution residuals for the chosen algo Fit for each algorithm the resolution, for the whole region and the BARREL ONLY?? PUT a plot C. Diez Pardos 11 11 Left tail: reconstructed momentum higher than generated Interested: See tails in low pt distributions and resolution core in high pt MEAN: Similar? SIGMA: Quite similar within algorithm, better for the chosen one
Resolution residuals for the chosen algo PONER % de picky y TPFMS Por ver que mejor no cogerlo nunca?? ==> Hacer el del cocktail total Results separated by selected or rejected for Picky and TPFMS Left tail: reconstructed momentum higher than generated Interested: See tails in low pt distributions and resolution core in high pt MEAN: Similar? SIGMA: Quite similar within algorithm, better for the chosen one C. Diez Pardos 12 12
Resolution residuals as a function of pt for W' MC Tracker not shown, too little stat C. Diez Pardos 13 13
Resolution residuals as a function of pt for W (allpt!) MC C. Diez Pardos 14 14
Pulls as a function of pt for W MC C. Diez Pardos 15 15
Pulls as a function of pt for W' MC C. Diez Pardos 16 16
Conclusions - How valid a tune with data/MC is applicable to the other samples: Different behaviour in eta regions and pt Datos: falta una contribucion para pt bajo de MC - Is it the optimal cocktail? (It seems that it works fine...) - Check with cosmics (Jordan?) Many, many thanks to Jordan for all the help and pacience C. Diez Pardos 17 17
Back-up C. Diez Pardos 18 18
Chi2 between data/MC (all BG) (Plots by G. Abendi, A. fanfani) C. Diez Pardos 19 19
Barrel Choice for MC and data
Data: all eta regions